 |
Step
1:
|
Identify a cohort of customers to be analyzed. Generally,
all customers in the cohort should have been acquired
within the same time period.
|
| |
Step
2: |
Create
the "cells" for the RFM analysis.
|
| |
Step
3: |
Create
an RFM matrix for the customer cohort.
|
| |
Step
4: |
Set
the number of purchase events for which the analysis will
be conducted (e.g., four future purchases).
|
| |
Step
5: |
Generate
all of the nodes of the decision tree. (The base of every
branch is a node; that is, every point of choice between
alternatives is a node.)
|
| |
Step
6: |
For
every node and terminal state, compute the probability
that the customer will reach it. (A terminal state marks
the end of a complete purchase path.)
|
| |
Step
7: |
Compute
the profitability of a customer at each terminal state.
|
| |
Step
8: |
Compute
the expected retention equity per customer by multiplying
the probability that a customer will reach a given terminal
state times the payout once the customer has reached it.
|
| The
example of Firm C will illustrate these steps. |
|
Step
1:
|
Identify
a cohort of customers to be analyzed. Generally, all customers
in the cohort should have been acquired within the same
time period. Firm C selected as its cohort 1,000 customers
acquired in 1995.
|
|
Step
2: |
Create the cells for RFM analysis. The cells must be mutually
exclusive, meaning that a customer has a clear path to
follow. In this case, the cells were based only on recency
and frequency, as shown in Table
Retention-8. With each first purchase, a path began
with the customer entering the one-time, 0-6 months cell.
From there, the customer became either a two-time buyer
with 0-6 months recency or a one-time buyer with 7-12
months recency.
Adding a monetary value component to the framework would
have increased the complexity of the analysis. For example,
if the firm had decided to add cells based on purchase
amount (e.g., purchases worth $0 to $100 versus those
worth more than $100), the number of cells in Table
Retention-8 would have doubled. It is important to
note that, depending on how firms define monetary value
in the RFM structure, a customer could increase the frequency
of purchases but still remain at the same level of monetary
value. For example, a customer could spend $25 on the
first purchase and $25 on the next purchase. Using the
monetary value breakdown noted above, this customer would
have a frequency of 2 but still remain in the monetary
value classification of $0-$100. Firms typically measure
monetary value over a specified time horizon. For example,
the monetary value variable might be defined as the total
monetary value of the customer's last year of purchases.
|
|
Step
3: |
Create
an RFM matrix for the customer cohort. The specific probabilities
for this example appear in Table
Retention-8. The probabilities represent the likelihood
that a customer will be in that cell at any given time.
|
|
Step
4: |
Set
the number of purchase events for which the analysis will
be conducted. Here, for simplicity's sake, the analysis
covers only two years. If this were a real-world analysis,
a computer program would run the computations for projections
of five to ten years.
|
|
Step
5: |
Generate all of the nodes of the decision tree. There
are 16 nodes in this example's decision tree, as shown
in Table
Retention-9.
|
|
Step
6: |
For
every node and terminal state, compute the probability
that the customer will reach it.
The probability that a customer will reach a particular
node, described as "current state" in Table
Retention-9, equals the probability that the customer
reached the prior node times the probability of the buying
decision made by the customer at that prior node. (Note
that Period 1 is the second purchase opportunity for the
customer: The first purchase opportunity occurred during
the acquisition period. The probability that a customer
will reach the current state of one purchase in the last
0-6 months is 1.0, because by definition the customer
has purchased once already.)
For example, looking at Table
Retention-9, there is a .2 probability that a customer
will buy in Period 1. (This corresponds to the buying
probability listed in the first, upper-left cell of the
RFM matrix in Table
Retention-8.) Thus, there remains a .8 probability
that the customer will not buy in Period 1. If the customer
does not buy, then he advances to the current state of
one purchase in the last 7-12 months, where there is a
.9 (.9=1-.a)
probability
that he will again not buy. If the customer again does
not buy, he advances to the current state of one purchase
in the last 13-18 months. Therefore, the probability that
a given customer will reach this state, one purchase in
the last 13-18 months, equals .8 times .9 = .72.
|
|
Step
7: |
Compute
the profitability of a customer at each terminal state.
To determine the profitability of each path, start by
counting up the number of purchases (i.e., buy decisions)
that the customer has made, not including the first purchase.
Multiply this purchase number by the dollar margin per
purchase, and then subtract the per-customer marketing
costs for each period. This gives the per customer profit
for that path.
|
|
Step
8: |
Compute
the expected retention equity per customer. This is done
by multiplying the probability that a customer will reach
a given terminal state by the payout once the customer
has reached it. For each terminal state, Table
Retention-9 shows the probability of reaching it along
with its profit per customer. Multiplying this profit
per customer by the probability yields an expected profit
for the terminal state. The sum of the expected profits
for all terminal states equals retention equity per customer.
That
value for this example is $1,035.44. |
